Back to list

Singing Voice Vibrato: Measurement and Modification

Authors Sciri, P.
Year 2011
Thesis Type Diploma thesis
Topic Audio Signal Processing
Keywords musical acoustics
Abstract The intention of this thesis is to present a concept to describe inherent attributes of the human singing voice in a qualitative manner. Many aspects of the singing voice are unique to it such as its wide tonal range, the rich repertoire of timbral colorizations or the possibilities regarding dynamics and phrasing while singing. The entity of all those attributes yields an instrument with very special properties. As one of those properties, vibrato can be singled out being one of the most artistic and virtuoso features of the singing voice. Therefore, this issue will be discussed prominently in the course of this thesis. Objective of this work will be to present a framework of measures to describe the singing voice on a physical basis considering the related anatomy and mechanisms involved in voice production. Based on state-of-the-art digital modeling techniques for the processes in the vocal apparatus the respective model parameters are determined from microphone recordings. The involved measurement procedures include pitch, amplitude and formant tracking. Special attention is paid to the estimation of the glottal source signal by the development of a new constrained closed phase glottal inverse filtering algorithm. The presented novelties in this work include the combination of multiple stages to determine the instant of glottal closure from microphone signals as well as cycle prototyping to ensure coherence of analysis. Exploiting the knowledge derived from a small scale empirical study on vocal vibrato, an algorithm for actively influencing vibrato in singing voice signals will be presented. A set of predefined vibrato patterns has been recorded by a classically educated baritone singer under studio conditions. These recordings are used to investigate the relationship between voice model parameters and the occurrence of vibrato. As one of the possible practical scenarios, vibrato cancellation will be discussed in detail. An assessment of conceptual and computational possibilities and limitations will be given. Furthermore, the usage of a linear time-variant Lattice filter as well as the implementation of an asymmetric pitch-synchronous overlap-and-add technique to perform synthesis will be presented. As one result, the complete reduction of a semitone vibrato to a static pitch is shown.
Supervisors Höldrich, R., Sontacchi, A.